Tim, please see inline ....
Werner --Original Message Text--- From: [EMAIL PROTECTED] Date: Wed, 4 May 2005 13:31:45 +0800 HI Werner, after a quick review of current and past Castor JDO issues in Jira for "caching", I stumbled across two entries : Add JMS publisher subscriber plugins for expire cache notification http://jira.codehaus.org/browse/CASTOR-875 Add a distributed cache implementation. http://jira.codehaus.org/browse/CASTOR-674 I have added the following comment to both issues : Is your JGroups-based Cache any closer to being integrated with Castor JDO 0.9.7 ? Can the code be checked into a "pending" folder of the Castor CVS tree for viewing ? Looking at CASTOR-875, he sample code from Kalpesh Patel is a workable custom JMS approach, but requires the web app to register a background listener (eg. a servlet) and is tied to JMS. Like yourself, I think a more abstract layer, using JGroups (or OSCache) is a better longer term approach and would bring Castor JDO closer to Hibernate, which already allows a pluggable distributed cache. werner --> I agree with you here, and most likely we will try to integrate cache implementations such as JCache and/or JCS. Personally, I have not looked at OSCache yet, but that might change as well. I have not yet attempted to code a custom approach (using a JGroups channel or OSCache cache), ... werner --> I'd really appreciate any help I could get with this, especially with testing once this has been coded up. Any code and time you are willing to donate would definitely be welcome. ... but am curious as to how the objects in the cache will be isolated for a transaction. werner --> Not that's a completely different issue. With a listener-based approach, I can definitely see why you are mentioning this issues. But with an in-memory cache distribution mechanism, most of these issues *might* be addressed the the underlying cache engine used. Consider the case where server A updates object A1 in the distributed cache, the cache framework fires a message to listeners on the network (via TCP, UDP or JMS) and server B then syncs its version of object A1. If object A is involved with transaction B1, how does the cache listener on server B update its version of object A1 without rolling back transaction B1 due to a timestamp mismatch (or equivalent). I think I'm missing something here with Castor JDO's caching and locking mechanics for a transaction. Comments welcome. Tim. +++ Tim Telcik Technical Application Architect Business Systems Delivery Information Technology Services, HBF, Inc. Email: [EMAIL PROTECTED] Tel: 9214 6744 Fax: 9214 6779 "Werner Guttmann" <[EMAIL PROTECTED]> 02/05/2005 06:04 PM Please respond to [email protected] To <[email protected]> cc Subject AW: [castor-dev] Distributed Castor JDO Cache Tim, no, nothing has been implemented yet. But we have been discussing this quite recently, and adding such distributed cache implemenations definitely is on our agenda for the one of the upcoming releases (0.9.9 or later). I've recently upgraded Castor's internal Cache interface so that integration with products like JCache or JCS will be easier. There's already an issue out there (assigned to myself) that will eventually add a distributed cache implementation that uses JCache internally. But I have decided to focus on support for polymorphism and Spring integration first, the former of which has been finished and awaits integration into the codebase. Given the way things go with the latter, I reckon that I'll be able to start working on such a cache implementation right after the 0.9.7 release in two weeks time Other than that, if you had some spare time (or some ready code .. ;-)), we'd really appreciate your contributions. Regards Werner -----Urspr�ngliche Nachricht----- Von: [EMAIL PROTECTED] [mailto:[EMAIL PROTECTED] Gesendet: Montag, 02. Mai 2005 11:51 An: [email protected] Cc: [EMAIL PROTECTED] Betreff: [castor-dev] Distributed Castor JDO Cache To whom it may concern, has anybody implemented a pluggable distributed (synchronized) cache for Castor JDO based on JCache (or a related distributed caching API) ? Bruce Snyder refers to the intention of using JCache in comments at "http://hypermail.linklord.com/castor- dev/2004/Jul/4697.html". Bruce's comments are in response to a query about distributed caches by Kalpesh Patel. We are looking at the problem of synchronizing multiple Castor JDO caches across multiple application servers on different hosts. A simple approach may be to notify listening Caches when the TxSynchronizable#commit method is invoked. JMS is a safer transport mechanism, but TCP is also a contender. Comments welcome. Tim. +++ Tim Telcik Technical Application Architect Business Systems Delivery Information Technology Services, HBF, Inc. Email: [EMAIL PROTECTED] Tel: 9214 6744 Fax: 9214 6779 ============================================================ HBF Health Funds Inc. a registered organisation under the National Health Act HBF Insurance Pty Ltd ACN 009 268 277 This email is a confidential communication intended only for the named Addressee(s). If you are not that person any disclosure, copying, distribution or use of this email or the information in it is prohibited. Please notify us immediately by telephone, (+61) 08 9265 6111 or return email and delete this email. ============================================================ HBF Health Funds Inc. a registered organisation under the National Health Act HBF Insurance Pty Ltd ACN 009 268 277 This email is a confidential communication intended only for the named Addressee(s). If you are not that person any disclosure, copying, distribution or use of this email or the information in it is prohibited. Please notify us immediately by telephone, (+61) 08 9265 6111 or return email and delete this email.
